Customer service intelligent interaction engine system driven by big models
Through the large-model-driven customer service intelligent interaction engine system, combined with the communication purpose harmony analysis and user similarity, screening data for large-model training, solving the problem that intelligent customer service cannot understand user personalized experience and complex context in the existing technology. , achieving more accurate user intention understanding and interactive optimization.
Patent Information
- Application Number
- CN202411080566.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-08
AI Technical Summary
The existing customer service intelligent interactive system cannot accurately understand the user's personalized experience and complex context, resulting in inaccurate responses or ambiguity, making it difficult to deal with practical problems in actual work.
Using a large model-driven customer service intelligent interaction engine system, the user's browsing and communication data is obtained through the data collection module, combined with communication purpose, harmony analysis and user similarity analysis, the data used for large model training is selected to optimize intelligent customer service. interactive effect.
It improves the efficiency of big model-driven, optimizes intelligent customer service interaction, and improves the accurate understanding of user intentions and the accuracy of replying.
Smart Images

Figure CN118608160B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a customer service intelligent interaction engine system driven by a large model. Background Art
[0002] Customer service is a bridge between enterprises and consumers, and its quality and efficiency are directly related to the brand image and market competitiveness of enterprises. With the rapid development of artificial intelligence technology, especially the breakthrough of large language models (LLMs), the customer service field urgently needs to be transformed into an intelligent one.
[0003] Existing technologies use the habitual patterns of groups to understand and respond to users' intentions according to preset rules and templates, and make mechanical replies. They do not pay attention to the personalized experience of users, have difficulty understanding users' ideas, and cannot cope with and solve various practical problems in actual work. In specific life scenarios, due to the wide variety of real situations, when dealing with complex and subtle contextual environments and semantics for individual users, smart customers still cannot accurately capture the true intentions of users, making customer service responses inaccurate or ambiguous. Summary of the invention
[0004] The present invention provides a customer service intelligent interaction engine system driven by a large model to solve the existing problems.
[0005] The customer service intelligent interaction engine system driven by a large model of the present invention adopts the following technical solutions:
[0006] An embodiment of the present invention provides a customer service intelligent interaction engine system driven by a large model, the system comprising the following modules:
[0007] A data collection module is used to obtain the browsing data and communication data of the user, wherein the browsing data at least includes: the name of the product browsed by the user, the number of product types, and the browsing time of each product; the communication data is the chat record generated when the user communicates with the intelligent customer service during the browsing process, and the chat record contains several communication processes, and the communication process contains several continuous sentences;
[0008] The communication purpose analysis module is used to obtain the user's communication purpose based on the difference in the time users spend browsing products;
[0009] The communication harmony analysis module is used to obtain the difference of consecutive sentences; perform word segmentation on the product names to obtain keywords, and obtain the communication harmony of consecutive sentences by integrating the occurrence of keywords in the communication process to which the consecutive sentences belong and the difference of the consecutive sentences; based on the communication harmony of the consecutive sentences, a communication harmony sequence of any communication process of the user is constructed; and the communication harmony of the communication process is obtained by combining the similarity of the communication harmony sequences of different communication processes and the communication purpose of the user;
[0010] The user similarity analysis module is used to obtain similar users of any user; analyze the similarity of the communication harmony between the user and similar users, and obtain the similarity between the user and similar users;
[0011] The large model training module is used to filter the communication data according to the similarity between the user and similar users, and to perform large model training on the browsing data and communication data of the filtered users.
[0012] Furthermore, the specific method of obtaining the user's communication purpose according to the user's browsing data includes:
[0013] Get the difference in the time users spend browsing products;
[0014] The communication purpose of the user is obtained by subtracting the difference between the number of product types the user browses and the length of time the user browses the product from the number of product types the user browses. The result is then subjected to an exponential function with a natural constant as the base, thus obtaining the difference.
[0015] Furthermore, the specific method for obtaining the difference in the time spent by users browsing products is as follows:
[0016] ;
[0017] in, Indicates the difference in the time users spend browsing products. Indicates that the user browses the product Length of time, Indicates browsing products Length of time, Indicates the number of products that the user browses. Represents the normalization function.
[0018] Furthermore, the specific method for obtaining the difference between consecutive sentences is as follows:
[0019] The distance editing algorithm is used to obtain the minimum number of edits between two consecutive sentences. The minimum number of edits between two consecutive sentences represents the difference between consecutive sentences, which is recorded as , indicating two consecutive sentences The difference of , Represents two consecutive sentences , The character length of
[0020] Furthermore, the word segmentation of the product name is performed to obtain keywords, and the communication harmony of the consecutive sentences is obtained by comprehensively considering the occurrence of keywords in the communication process of the consecutive sentences and the difference of the consecutive sentences. The specific method includes:
[0021] Jieba algorithm is used to segment the product names to obtain several phrases, which are recorded as keywords of the corresponding products. The number of times the keywords appear in each communication process is counted.
[0022] The method for obtaining the communication harmony of any two consecutive sentences is:
[0023] ;
[0024] in, Indicates a continuous statement Harmony of communication; Indicates a continuous statement The number of times the key words appear in the communication process; Indicates a continuous statement The number of consecutive sentences in the communication process; Indicates a continuous statement The difference of , Represents consecutive sentences , The character length of is an exponential function with a natural constant as its base.
[0025] Furthermore, the specific method for obtaining the communication harmony degree sequence of any communication process of the user based on the communication harmony degree of consecutive sentences is as follows:
[0026] The communication harmony of all consecutive sentences in any communication process of the user is sorted according to the time sequence in the chat record to form a communication harmony sequence in any communication process of the user.
[0027] Furthermore, the communication harmony degree of the communication process is obtained by combining the similarity of the communication harmony degree sequences of different communication processes and the communication purpose of the user, and the specific method includes:
[0028] Using a dynamic time warping algorithm to obtain the DTW distance of the communication harmony degree sequences of any two communication processes as the similarity of the communication harmony degree sequences of the two communication processes;
[0029] The method for obtaining the communication harmony of any communication process is:
[0030] ;
[0031] in, Indicates The communication harmony of the communication process, Indicates The communication purpose of the user in this communication process, Indicates The communication harmony sequence of the communication process, Indicates The communication harmony sequence of the communication process, represents the number of communication processes, Indicates The second communication process The similarity of the communication harmony sequences of the communication processes.
[0032] Furthermore, the specific method of obtaining similar users of any user includes:
[0033] All keywords in the communication process are converted into word vectors through the Word2vec model, and all word vectors are clustered using the K-means algorithm to obtain several clusters, and several users corresponding to all word vectors in any cluster are obtained; for any user in any cluster, the remaining users in the cluster to which the user belongs are recorded as similar users of the user.
[0034] Furthermore, the similarity of the communication harmony between the user and similar users is analyzed to obtain the similarity between the user and similar users, including the specific method of:
[0035] The method for obtaining the similarity between any user and similar users is:
[0036] ;
[0037] in, Indicates user Similarity with similar users, Indicates user The average value of the communication harmony of all communication processes, Indicates user The average value of the communication harmony of all communication processes of all similar users, is an exponential function with a natural constant as base, represents the absolute value function.
[0038] Furthermore, the communication data is screened according to the similarity between the user and similar users, and a large model is trained on the browsing data and communication data of the screened users, including the specific method of:
[0039] A first threshold A is preset, and users whose similarity with similar users is greater than or equal to A are recorded as target users. The browsing data and communication data of the target users are stored as data sets for training, and a large language model is trained. The trained large language model is used as a customer service interaction engine to realize online consulting services for customers.
[0040] The beneficial effects of the technical solution of the present invention are as follows: the present invention determines the communication purpose between the user and the intelligent customer service by analyzing the purpose of the user during the browsing process, determines whether the communication between the user and the intelligent customer service is harmonious based on the communication effect between the user and the intelligent customer service and in combination with historical communication records, determines the similarity between similar users based on the communication effects of similar user groups, and determines the communication data ultimately used for large model driving, thereby improving the efficiency of large model driving and facilitating the optimization of intelligent customer service interaction. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0042] Figure 1 This is a structural block diagram of the customer service intelligent interaction engine system driven by the large model of the present invention. DETAILED DESCRIPTION
[0043] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of the customer service intelligent interaction engine system driven by the large model proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0044] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0045] The specific scheme of the customer service intelligent interaction engine system driven by a large model provided by the present invention is described in detail below with reference to the accompanying drawings.
[0046] See also Figure 1 , which shows a structural block diagram of a customer service intelligent interaction engine system driven by a large model provided by an embodiment of the present invention, the system includes the following modules:
[0047] The data collection module is used to obtain users' browsing data and communication data.
[0048] In order to implement the large model-driven customer service intelligent interaction engine system proposed in this embodiment, it is first necessary to collect the user's browsing data and communication data. The specific process is as follows:
[0049] The browsing data and communication data of several users on the shopping platform during several browsing processes are obtained, wherein the browsing data includes the name of the browsed products, the number of products and the number of product types, and the browsing time of each product; the communication data is the chat record generated when the user communicates with the intelligent customer service during the browsing process, and the chat record contains several communication processes, and the communication process contains several continuous sentences.
[0050] It should be noted that, in this embodiment, when the duration of previous and subsequent communications in a chat record exceeds a preset duration T, the previous chat content is regarded as one communication process.
[0051] It should be noted that T is preset to 5 minutes based on experience, and can be adjusted according to actual conditions, and is not specifically limited in this embodiment.
[0052] So far, the user's browsing data and communication data are obtained through the above method.
[0053] The communication purpose analysis module is used to obtain the user's communication purpose based on the user's browsing data.
[0054] It should be noted that the existing customer service intelligent interaction engine system, that is, the intelligent system of intelligent customer service, often has unsatisfactory effects in real life. Taking e-commerce customer service as an example, its main functions are order query, transaction record, and appeal processing, etc. However, in real life scenarios, because each person has different living habits, different ways of expression, and different demands for expression, its intelligent customer service often finds it difficult to understand the meaning expressed by each user, and cannot make effective answers, or solve practical problems, and cannot achieve the purpose of interaction well. It cannot understand the user's language. The essential reason is that my country's language is rich, and there are many ways to express the same meaning. In daily life scenarios, people's language expression tends to be more colloquial, not like official document writing, with a rigorous subject-predicate-object structure and clear logic. Different people's expression habits will also be different, such as regional differences, group differences, etc. Therefore, the present invention analyzes the interactive language characteristics of users through the analysis of the characteristics of the user's personal behavior habits and expression methods, and determines whether the user's current language is emotional language or event language, and makes corresponding countermeasures for different characteristics.
[0055] It should be noted that, based on the user's behavioral habits on the platform, we can make a preliminary judgment on the user's language style, judge the current contextual environment of the conversation with the user, and obtain the user's browsing data on the online trading platform. When the user browses the product information casually on the online trading platform, just follows the recommendations on the homepage and views the products aimlessly. There is no specific search or viewing of a certain category of products, and no obvious inclination. He is just browsing the products aimlessly. At this time, the user does not have a strong purpose and is just killing time. If the user has a conversation with the intelligent customer service at this time, the user's language style should be an emotional style, and the user's communication purpose is relatively weak.
[0056] It should be noted that if the user's browsing time for different products is similar, it means that the user's browsing purpose is weak, and the communication purpose is weaker. The more categories of products browsed at the same time, the weaker the user's communication purpose.
[0057] Specifically, first, obtain the difference in the time that users spend browsing products.
[0058] As an embodiment, a method for obtaining the difference in the time spent by users browsing a product is as follows:
[0059] ;
[0060] in, Indicates the difference in the time users spend browsing products. Indicates that the user browses the product Length of time, Indicates browsing products Length of time, Indicates the number of products that the user browses. Represents the normalization function.
[0061] Secondly, obtain the user’s communication purpose.
[0062] As an embodiment, a method for obtaining the user's communication purpose is:
[0063] ;
[0064] in, Indicates the user's communication purpose. Indicates the number of product categories that the user browses. Indicates the difference in the time users spend browsing products. is an exponential function with a natural constant as its base.
[0065] It should be noted that the embodiment shows that there is no obvious difference in the time a user spends browsing different products during a single browsing process. Specifically, the browsing time of different products during a single browsing process is similar, that is,
[0066] The smaller it is, the more likely it is that users do not have a clear tendency to browse or select a certain type of product during the product browsing process. The larger the value, the more miscellaneous the products browsed by the browsing user are, without obvious characteristics, and the communication purpose of the user is weak. For example, a user browsed products of different categories on the platform at a certain time, and the time difference between the products he browsed was small, so the communication purpose of the user was weak.
[0067] So far, the user's communication purpose is obtained through the above method.
[0068] The communication harmony analysis module is used to obtain the difference between consecutive sentences; segment the product names to obtain keywords, and comprehensively consider the occurrence of keywords in the communication process to which the consecutive sentences belong and the difference between the consecutive sentences to obtain the communication harmony of the consecutive sentences; based on the communication harmony of the consecutive sentences, a communication harmony sequence of any communication process of the user is constructed; and the communication harmony of the communication process is obtained by combining the similarity of the communication harmony sequences of different communication processes and the communication purpose of the user.
[0069] It should be noted that the communication purpose analysis module calculates the communication purpose between the user and the intelligent customer service through quantitative analysis of the user's behavioral habits of browsing products on the platform, and determines the communication scenario when the user communicates with the customer service. Therefore, this step analyzes the coherence and progressiveness of the user's language during the conversation with the intelligent customer service based on the communication scenario when the user currently communicates with the intelligent customer service.
[0070] It should be noted that when communicating with users in the system, when there is obvious coherence between the user's previous and next sentences, it indicates that the user is relatively satisfied with the current reply of the intelligent customer service, or it can arouse the user's interest, so that the user further asks questions on the current reply and believes that the current reply is a valid reply, that is, the reply is an accurate understanding of the user's language.
[0071] It should be noted that in the communication between users and intelligent customer service, the communication harmony of the communication process is related to the effectiveness of the communication content. The larger the proportion of effective communication, the more effective the communication content, and therefore the higher the communication harmony of the communication process; at the same time, the higher the frequency of keywords in the communication process, the more relevant the communication content is to the user's needs, and the more effective the communication process is, the higher the communication harmony of the communication process; in addition, the more the correlation of sentences in user communication reflects the user's communication harmony, the smaller the difference between consecutive sentences, the more coherent the user communication is, and the more harmonious the communication is at this time, and therefore the higher the communication harmony of the communication process.
[0072] Specifically, first, the difference between consecutive sentences is obtained.
[0073] The distance editing algorithm is used to obtain the minimum number of edits between two consecutive sentences. The minimum number of edits between two consecutive sentences represents the difference between consecutive sentences, which is recorded as , indicating two consecutive sentences The difference of , Represents two consecutive sentences , The character length of
[0074] It should be noted that the distance editing algorithm is an existing method for measuring the similarity between two character strings, so it is not described in detail in this embodiment.
[0075] Secondly, the product names are segmented to obtain keywords, and the communication harmony of consecutive sentences is obtained by comprehensively considering the occurrence of keywords in the communication process of consecutive sentences and the differences of consecutive sentences.
[0076] Jieba algorithm is used to segment the product names to obtain several phrases, which are recorded as keywords of the corresponding products. The number of times the keywords appear in each communication process is counted.
[0077] It should be noted that, because in the online shopping platform scenario, the communication between users and intelligent customer service is mostly inquiries or problem solving, there will be obvious product-related vocabulary, and the keywords also represent product-related vocabulary. Therefore, the frequency of keyword appearance during a single communication process can be analyzed to analyze the user's communication harmony.
[0078] It should be noted that the Jieba algorithm is an existing Chinese word segmentation algorithm that combines a word segmentation method based on string matching and a word segmentation method based on statistics, so it will not be described in detail in this embodiment.
[0079] As an embodiment, the method for obtaining the communication harmony of any two consecutive sentences is:
[0080] ;
[0081] in, Indicates a continuous statement Harmony of communication; Indicates a continuous statement The number of times the key words appear in the communication process; Indicates a continuous statement The number of consecutive sentences in the communication process; Indicates a continuous statement The difference of , Represents consecutive sentences , The character length of is an exponential function with a natural constant as its base.
[0082] It should be noted that It indicates the proportion of keywords appearing during any communication process of the user. The more keywords appear, the more effective the response is, and the higher the communication harmony is. When there are more effective responses and the user always communicates around keywords during the communication and response process, that is, the smaller the difference between consecutive sentences is, the better the communication harmony between the user and the intelligent customer service is.
[0083] Then, based on the communication harmony of consecutive sentences, a communication harmony sequence of any communication process of the user is constructed.
[0084] The communication harmony of all consecutive sentences in any communication process of the user is sorted according to the time sequence in the chat record to form a communication harmony sequence in any communication process of the user.
[0085] Finally, the communication harmony of the communication process is obtained by combining the similarity of the communication harmony sequences of different communication processes and the communication purpose of the users.
[0086] It should be noted that there is a deviation in judging the communication harmony only by the user's continuous sentences. The communication harmony of the communication records should also be obtained by combining multiple communication records of the intelligent customer service. If multiple communication records indicate that the communication harmony of the communication records is relatively similar, and the user has a higher communication purpose, then it can be said that the user's communication harmony is higher, that is, the user's communication harmony is higher after adjustment.
[0087] The dynamic time warping algorithm is used to obtain the DTW distance of the communication harmony degree sequences of any two communication processes as the similarity of the communication harmony degree sequences of the two communication processes.
[0088] It should be noted that the DTW (Dynamic Time Warping) algorithm is an existing algorithm for dynamic time planning, which can be used to compare the similarity of two sequences, so it is not described in detail in this embodiment.
[0089] As an embodiment, a method for obtaining the communication harmony degree of any communication process is as follows:
[0090] ;
[0091] in, Indicates The communication harmony of the communication process, Indicates The communication purpose of the user in this communication process, Indicates The communication harmony sequence of the communication process, Indicates The communication harmony sequence of the communication process, represents the number of communication processes, Indicates The second communication process The similarity of the communication harmony sequences of the communication processes.
[0092] It should be noted that when the user The greater the similarity between the communication harmony degree of this communication process and the communication harmony degree of other historical communication processes between users and intelligent customer service, that is, The larger the The communication process between the first user and the intelligent customer service is similar to the second The more similar the communication processes are, the more similar the communication harmony sequence between the current communication process between the customer and the intelligent customer service is with each communication process in history is calculated. The greater the similarity between the current communication process and the historical communication process, the clearer the intelligent customer service understands the user's statement. At the same time, the stronger the user's purpose and the more similar it is to the historical communication records, the higher the harmony of the interaction between the user and the intelligent customer service is.
[0093] The user similarity analysis module is used to obtain similar users of any user; analyze the similarity of the communication harmony between the user and the similar users, and obtain the similarity between the user and the similar users.
[0094] It should be noted that, because of the interactive scenario of the intelligent customer service based on the online shopping platform, when a certain keyword appears multiple times during the interactive communication process, in this scenario, based on the conversation content between other users and the intelligent customer service, the communication between this user and the intelligent customer service has certain similarities. For example, if a user talks about a certain category of goods, then when the system detects the keyword, the current target user of the category of goods and the group that purchases the goods may ask the same question. Therefore, it can be obtained through
[0095] Specifically, all keywords in the communication process are converted into word vectors through the Word2vec model, and all word vectors are clustered using the K-means algorithm to obtain several clusters, and several users corresponding to all word vectors in any cluster are obtained; for any user in any cluster, the remaining users in the cluster to which the user belongs are recorded as similar users of the user.
[0096] It should be noted that the Word2vec model is an existing model for generating word vectors, so it is not described in detail in this embodiment.
[0097] It should be noted that the K-means algorithm is an existing clustering algorithm, so it is not described in detail in this embodiment.
[0098] It should be noted that, in this embodiment, the K value is preset to 4, which can be adjusted according to actual conditions and is not specifically limited in this embodiment.
[0099] As an embodiment, a method for obtaining the similarity between any user and similar users is as follows:
[0100] ;
[0101] in, Indicates user Similarity with similar users, Indicates user The average value of the communication harmony of all communication processes, Indicates user The average value of the communication harmony of all communication processes of all similar users, is an exponential function with a natural constant as base, represents the absolute value function.
[0102] It should be noted that when the user The more similar the communication harmony is with similar users, the more similar the current intelligent customer service is with the user. The better the communication experience, the better the understanding of the user. The higher the similarity with similar users.
[0103] So far, the similarity between any user and similar users is obtained through the above method.
[0104] The large model training module is used to filter the communication data according to the similarity between the user and similar users, and to perform large model training on the browsing data and communication data of the filtered users.
[0105] Specifically, a first threshold A is preset, and users whose similarity with similar users is greater than or equal to A are recorded as target users. The browsing data and communication data of the target users are stored as data sets for training, and a large language model is trained. The trained large language model is used as a customer service interaction engine to realize online consulting services for customers.
[0106] It should be noted that A is preset to 0.7 based on experience and can be adjusted according to actual conditions, and is not specifically limited in this embodiment.
[0107] At this point, this embodiment is completed.
[0108] It should be noted that the The model is only used to represent negative correlation and constrain the output of the model to be in In the specific implementation, it can be replaced by other models with the same purpose. This embodiment is only based on The model is described as an example without any specific limitation. is the input to the model.
[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. Customer service intelligent interaction engine system driven by large models, It is characterized in that The system includes the following modules: A data collection module is used to obtain the browsing data and communication data of the user, wherein the browsing data at least includes: the name of the product browsed by the user, the number of product types, and the browsing time of each product; the communication data is the chat record generated when the user communicates with the intelligent customer service during the browsing process, and the chat record contains several communication processes, and the communication process contains several continuous sentences; The communication purpose analysis module is used to obtain the user's communication purpose based on the difference in the time users spend browsing products; The communication harmony analysis module is used to obtain the difference of consecutive sentences; perform word segmentation on the product names to obtain keywords, and obtain the communication harmony of consecutive sentences by integrating the occurrence of keywords in the communication process to which the consecutive sentences belong and the difference of the consecutive sentences; based on the communication harmony of the consecutive sentences, a communication harmony sequence of any communication process of the user is constructed; and the communication harmony of the communication process is obtained by combining the similarity of the communication harmony sequences of different communication processes and the communication purpose of the user; The user similarity analysis module is used to obtain similar users of any user; analyze the similarity of the communication harmony between the user and similar users, and obtain the similarity between the user and similar users; The large model training module is used to filter the communication data according to the similarity between the user and similar users, and to train the large model on the browsing data and communication data of the filtered users; The specific method of obtaining the user's communication purpose according to the user's browsing data is as follows: Get the difference in the time users spend browsing products; The number of product categories browsed by the user minus the product of the difference between the number of product categories browsed by the user and the time the user browsed, and the result is subjected to an exponential function with a natural constant as the base to obtain the user's communication purpose; The method for obtaining the communication purpose is: ; in, Indicates the user's communication purpose. Indicates the number of product categories that the user browses. Indicates the difference in the time users spend browsing products. is an exponential function with a natural constant as base; The specific method for obtaining the difference between consecutive sentences is as follows: The distance editing algorithm is used to obtain the minimum number of edits between two consecutive sentences. The minimum number of edits between two consecutive sentences represents the difference between consecutive sentences, which is recorded as , indicating two consecutive sentences , The difference of , Represents two consecutive sentences , The character length of The method of performing word segmentation on the product name to obtain keywords, comprehensively considering the occurrence of keywords in the communication process of the consecutive sentences and the difference of the consecutive sentences, and obtaining the communication harmony of the consecutive sentences includes the following specific methods: Jieba algorithm is used to segment the product names to obtain several phrases, which are recorded as keywords of the corresponding products. The number of times the keywords appear in each communication process is counted. The method for obtaining the communication harmony of any two consecutive sentences is: ; in, Indicates a continuous statement , Harmony of communication; Indicates a continuous statement , The number of times the key words appear in the communication process; Indicates a continuous statement , The number of consecutive sentences in the communication process; Indicates a continuous statement , The difference of , Represents consecutive sentences , The character length of is an exponential function with a natural constant as its base.
2. According to the large model-driven customer service intelligent interaction engine system of claim 1, It is characterized in that The specific method for obtaining the difference in the time users spend browsing products is as follows: ; in, Indicates the difference in the time users spend browsing products. Indicates that the user browses the product Length of time, Indicates browsing products Length of time, Indicates the number of products that the user browses. Represents the normalization function.
3. According to the large model-driven customer service intelligent interaction engine system of claim 1, It is characterized in that The specific method for obtaining the communication harmony degree sequence of any communication process of the user based on the communication harmony degree of continuous sentences is as follows: The communication harmony of all consecutive sentences in any communication process of the user is sorted according to the time sequence in the chat record to form a communication harmony sequence in any communication process of the user.
4. According to the large model-driven customer service intelligent interaction engine system of claim 1, It is characterized in that The specific method of combining the similarity of the communication harmony degree sequences of different communication processes and the communication purpose of the user to obtain the communication harmony degree of the communication process includes: Using a dynamic time warping algorithm to obtain the DTW distance of the communication harmony degree sequences of any two communication processes as the similarity of the communication harmony degree sequences of the two communication processes; The method for obtaining the communication harmony of any communication process is: ; in, Indicates The communication harmony of the communication process, Indicates The communication purpose of the user in this communication process, Indicates The communication harmony sequence of the communication process, Indicates The communication harmony sequence of the communication process, represents the number of communication processes, Indicates The second communication process The similarity of the communication harmony sequences of the communication processes.
5. According to the large model-driven customer service intelligent interaction engine system of claim 1, It is characterized in that The specific method of obtaining similar users of any user includes: All keywords in the communication process are converted into word vectors through the Word2vec model, and all word vectors are clustered using the K-means algorithm to obtain several clusters, and several users corresponding to all word vectors in any cluster are obtained; for any user in any cluster, the remaining users in the cluster to which the user belongs are recorded as similar users of the user.
6. According to the large model-driven customer service intelligent interaction engine system of claim 1, It is characterized in that The specific method of analyzing the similarity of the communication harmony between the user and similar users to obtain the similarity between the user and similar users is as follows: The method for obtaining the similarity between any user and similar users is: ; in, Indicates user Similarity with similar users, Indicates user The average value of the communication harmony of all communication processes, Indicates user The average value of the communication harmony of all communication processes of all similar users, is an exponential function with a natural constant as base, represents the absolute value function.
7. According to the large model-driven customer service intelligent interaction engine system of claim 1, It is characterized in that The communication data is screened according to the similarity between the user and similar users, and the large model training is performed on the browsing data and communication data of the screened users, including the specific method of: A first threshold A is preset, and users whose similarity with similar users is greater than or equal to A are recorded as target users. The browsing data and communication data of the target users are stored as data sets for training, and a large language model is trained. The trained large language model is used as a customer service interaction engine to realize online consulting services for customers.
Citation Information
Patent Citations
Customer service staff intelligent assistance method and device, server and storage medium
CN108920715A
Session method, and method and equipment for obtaining link information associated with object
CN110011895A
Intelligent customer service response method and system
CN110162611A
Commodity data processing method, device and equipment
CN111984837A
Answer generation method and device based on artificial intelligence, equipment and storage medium
CN113239169A
